Q-Learning-based-routing-protocol

Introduction

Let us consider N drones ${d_0, ..., d_{N−1}}$ deployed in an Area of Interest (AoI). Each drone $d_i$ has a mission assigned which consists of following a trajectory in the AoI and capturing events. Such events generate packets that have to be sent to the depot. Once a packet has been generated, the drone, can either keep it in its buffer until it reaches the depot or finds a drone to use as a relay and transmit all the packets in its buffer to the relay. Packets can also expire and have a deadline to be delivered, therefore our goal is to deliver them as quickly as possible to the depot.

Approaches

To solve the routing protocol the Q-Learning algorithm was used which is an off-policy TD control algorithm in Reinforcement Learning:

• greedy (with exploration in the early stages)
• greedy (without exploratin in the early stages)
• best action (with exploration in the early stages)
• best action (without exploration in the early stages)
• Q-FANET

The simulator used for the experiments can be found at this link https://github.com/flaat/DroNETworkSimulator/

If you want to try the solutions, you can put the routing algorithms in this folder https://github.com/flaat/DroNETworkSimulator/tree/main/src/routing_algorithms